We started on optimization (mostly through vectorization) of the hybrid operations throughout the library, ie using internal numpy vectors for storing sparse vectors inside the SparseEmbedding objects, more numpy in LocalHybridIndex, etc — but there are likely still places where we may be unecessarily moving from numpy to python and back again, or not fully utilizing numpy — could we work on these and get everything better optimized for hybrid?
We started on optimization (mostly through vectorization) of the hybrid operations throughout the library, ie using internal numpy vectors for storing sparse vectors inside the
SparseEmbeddingobjects, more numpy inLocalHybridIndex, etc — but there are likely still places where we may be unecessarily moving from numpy to python and back again, or not fully utilizing numpy — could we work on these and get everything better optimized for hybrid?